LG AI Research is best understood as an AI research and model ecosystem, not as a single everyday chatbot app. Its EXAONE family includes language models, vision-language models that can understand images and documents, lightweight models for local or on-device use, and specialized systems for enterprise and industrial work. Some models and demonstrations are available free for eligible research, academic, and educational use, while commercial access depends on model licensing, enterprise arrangements, or partner infrastructure.
What is LG AI Research?
LG AI Research is LG's dedicated artificial intelligence research organization, established in December 2020. It develops AI models and applied systems for language, images, documents, science, manufacturing, materials, biotechnology, and other professional areas.
For most users, the most relevant part of its work is the EXAONE model family. EXAONE includes general language models, reasoning models, vision-language models, smaller on-device models, and systems designed for enterprise or industrial use. A vision-language model is an AI system that can work with both written instructions and visual information, such as photographs, charts, scans, and documents.
LG AI Research also provides an EXAONE Showroom and publishes selected models through official or affiliated Hugging Face repositories. These resources make parts of the ecosystem accessible to the public, but they should not be confused with a fully developed consumer service such as ChatGPT, Gemini, or Claude.
What can an ordinary user use it for?
Public EXAONE demonstrations and model releases can be useful for trying AI-assisted work and understanding LG's technology. Depending on the specific model or interface, practical uses include:
- Asking questions and drafting text: language models can help explain a topic, summarize information, brainstorm ideas, or produce a first draft.
- Writing and rewriting: users can request clearer wording, translations, shorter versions, or changes in tone for suitable text-based tasks.
- Studying and research: reasoning-oriented models can help organize information, work through questions, and compare possible answers. Their responses should still be checked, especially for important decisions.
- Working with documents: multimodal EXAONE models are designed to understand documents and visual content, which can help with extracting or interpreting information from pages, tables, and images.
- Coding: EXAONE releases include coding capabilities that can help explain code, suggest implementations, or identify possible errors.
- Visual reasoning: supported vision-language models can interpret images and connect visual information with a written request.
The exact experience depends on the model and the interface being used. LG AI Research's ecosystem is not presented as one uniform application with the same features everywhere.
Is LG AI Research free?
There is no conventional consumer subscription plan clearly documented for LG AI Research. Instead, free access is available in several forms. Selected open-weight EXAONE models can be downloaded or used by eligible researchers, academics, and educators under the applicable model licenses. The EXAONE Showroom and selected Hugging Face demonstrations also provide public ways to explore the technology.
Open-weight means that model files are released for permitted users to inspect, run, or adapt under stated conditions. It does not automatically mean unrestricted commercial use. Each model may have its own license, and commercial deployment may require additional review or a separate agreement.
LG AI Research has announced commercial API access for EXAONE 4.0 through a partnership with FriendliAI. An API is a way for software developers to connect their own applications to an AI model. However, current public information supplied for this page does not verify a complete price list, quota system, authentication process, SDK offering, or a broad self-service first-party API for all current models.
In practical terms, users should expect:
- free public demonstrations and selected research access;
- model-specific terms for downloading and running open-weight releases;
- commercial access that may involve licensing, enterprise contracts, on-premise deployment, or a partner service;
- no clearly documented monthly consumer tiers comparable to ChatGPT Plus or Claude Pro.
How do you access it?
The simplest starting point is the official LG AI Research website and its EXAONE pages. These provide information about current releases, demonstrations, research announcements, and links to model repositories. A public EXAONE Showroom is intended for trying selected capabilities through a web interface, although availability and features may change.
Technical users can review the official Hugging Face and GitHub repositories for eligible open-weight models. Running a model locally generally requires suitable computer hardware, software setup, and compliance with the model's license. Lightweight variants are intended to make local or on-device deployment more practical, while larger models may require more substantial infrastructure.
Businesses should treat access as a separate evaluation. They may need to contact LG AI Research or an infrastructure partner to confirm model availability, commercial rights, security arrangements, deployment options, and support. The former ChatEXAONE public beta should not be treated as a current general-purpose chatbot: the supplied research states that the beta concluded in 2025.
What are the important EXAONE capabilities?
Language and reasoning
EXAONE models are designed for text generation, multilingual work, coding, and reasoning. EXAONE 4.0 introduced a hybrid approach with reasoning and non-reasoning modes. A reasoning mode is intended for tasks that benefit from more deliberate problem-solving, while a non-reasoning mode can be used when a quicker, more direct response is preferable.
The documented language coverage for EXAONE 4.0 includes Korean, English, and Spanish. Korean-language capability is a particularly important reason to consider the ecosystem, especially for organizations that need AI designed with Korean professional and business use in mind.
Images and documents
EXAONE 4.5 is identified by LG AI Research as its first publicly released open-weight vision-language model. It supports text and image understanding, document analysis, visual reasoning, coding, and tool-use-related capabilities. For a user, this means the model can potentially combine a written question with information found in an image or document rather than handling text alone.
Document understanding can be useful for tasks such as examining pages, extracting relevant details, or asking questions about visual material. Results should be checked when the source is complex, poorly scanned, sensitive, or important for legal, financial, medical, or operational decisions.
Coding, tools, and agents
EXAONE 4.0 includes support for function calling and the Model Context Protocol, or MCP. Function calling allows a model to request a specific software function rather than only returning conversational text. MCP is a way of connecting AI systems with external tools and data sources through a common protocol. These capabilities are mainly relevant to developers and organizations building workflows, agents, or internal applications.
LG AI Research also describes agent-focused work. In this context, an AI agent is a system that can plan or carry out multiple steps using tools, rather than simply answering one prompt. The existence of these capabilities does not mean that every public demonstration offers autonomous agents or unrestricted connections to outside services.
Models and deployment choices
The EXAONE ecosystem includes different model sizes and purposes rather than one model intended for every situation. EXAONE 4.0 included a 32B professional model and a 1.2B on-device variant. The smaller variant is designed for environments where running AI closer to the user or device matters.
LG AI Research's broader work includes language intelligence, physical intelligence, bio intelligence, data intelligence, materials intelligence, and advanced agents. These labels reflect research and application areas, not necessarily separate consumer products that anyone can open and use today.
Organizations may consider several deployment approaches:
- Web demonstrations: convenient for exploring selected features without managing infrastructure.
- Open-weight local inference: running an eligible model on an organization's own hardware or cloud environment.
- On-device deployment: using a smaller model locally to reduce dependence on remote servers.
- On-premise deployment: operating AI within an organization's own environment for control, security, or operational reasons.
- Hosted commercial access: using partner-supported infrastructure such as the announced FriendliAI route for EXAONE 4.0.
Strengths and limitations
Main strengths
- Strong emphasis on Korean-language and professional-domain AI.
- Public open-weight releases that give researchers and developers more deployment flexibility than a closed consumer-only service.
- Support for both reasoning and non-reasoning workflows in the EXAONE 4.0 generation.
- Multimodal and document-understanding research, including the open-weight EXAONE 4.5 vision-language model.
- Smaller on-device models and options for local or on-premise use.
- Research and solutions aimed at manufacturing, science, materials, biotechnology, and other enterprise settings.
Important limitations
- It is not currently presented as a mature, all-purpose consumer chatbot with a clearly documented subscription structure.
- Public demos may be experimental, limited, or subject to change.
- Open-weight availability does not guarantee unrestricted commercial rights; licenses must be checked model by model.
- Commercial API access is partner-mediated rather than documented as a broad, self-service first-party developer platform.
- Public information about pricing, quotas, authentication, SDKs, and current API model availability is incomplete.
- Compared with major global consumer AI services, it has fewer clearly documented applications, mobile or desktop products, and consumer-oriented integrations.
Privacy and data considerations
LG AI Research emphasizes on-device EXAONE deployment as a way to reduce server dependence and improve security and personal-data protection. Local or on-premise operation can be valuable for organizations that need greater control over sensitive information.
That benefit depends on the deployment and service being used. A detailed, current public summary of how prompts or submitted data from the EXAONE Showroom or commercial API are used for model training was not verified in the supplied research. Users should review the privacy policy, model license, partner terms, and enterprise agreement before submitting confidential information.
Who should consider LG AI Research?
LG AI Research is a good fit for researchers, developers, universities, and businesses investigating open-weight models, Korean-language AI, document understanding, local inference, or industry-specific applications. It may also suit technically capable users who want to experiment with an openly released model rather than use only a hosted consumer assistant.
It is less suitable for someone who simply wants a polished everyday chatbot with a well-known monthly plan, mobile applications, integrated web search, broad third-party integrations, and a large consumer support ecosystem. That user may find a mainstream consumer AI assistant easier to start with.
Practical assessment: choose LG AI Research for its EXAONE models, Korean and multilingual capability, multimodal document work, open-weight releases, and local or enterprise deployment possibilities. Its main drawbacks are the fragmented access model, limited consumer productization, uncertain public API details, and model-specific licensing. A competing consumer chatbot may make more sense for casual everyday use, while another enterprise or open-model provider may be preferable when transparent pricing, mature SDKs, or a large application ecosystem is the priority.